Skip to content

Set up your AI

You only need one AI coding tool. sdd installs a few small files for that tool and keeps the real rules in one place.

Your editor (VS Code, Cursor, …) is not the AI.
Your AI (Copilot, Grok, Claude, Ollama) is what writes with you.


Just set it up (this is enough)

bash
# In your app folder — pick ONE:
sdd init --here --ai copilot    # GitHub Copilot
# sdd init --here --ai grok
# sdd init --here --ai claude
# sdd init --here --ai ollama   # then: ollama pull llama3.2

sdd doctor
sdd new "My first change" -w hotfix -y

That’s it for day one. The AI often opens after process commands so it can help with the current step.

Learn without the AI opening? Add --no-agent (see tutorial).

Switch AI later:

bash
sdd agents install --ai claude --force

Which AI should I pick?

You already use…Choose
Copilot in VS Code / Cursorcopilot
Grok Build CLIgrok
Claude Code CLIclaude
Local models onlyollama

Full host details: Available agents.


What happens when I run commands?

After this…Does the AI open?
sdd new, next, verify, complete, agent, …Yes (unless --no-agent)
sdd status, doctor, workflows, feature listNo
AIHow it runs
grok / claude / ollamaTerminal launches the CLI
copilotOpen Copilot Chat → agent sdd (handoff file is refreshed)

Skip the AI once: --no-agent or SDD_NO_AGENT=1.


Where do the rules live?

You don’t need to memorize this on day one.

FileRole
.sdd/protocol.mdThe one short playbook
.sdd/active-context.mdWhat you’re doing right now
.sdd/handoff.mdBrief the AI just saw
Host folder (.github/agents, .grok/rules, …)Tiny pointers to the playbook

One host at a time. Installing a new AI removes the other hosts’ agent folders so you don’t get a mess of stubs.


Optional: roles (Copilot / Claude)

Some hosts get small role files (sdd, sdd-planner, sdd-implementer, sdd-reviewer).
They only say “read the live context, then the playbook.” Grok and Ollama use a single short brief instead.